Mathematical Reasoning on AMC 2023 (Avg@K Metrics)
86.75Avg@32PPPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| PPPOBackbone=Qwen3-8B, Evaluation Protocol=zero-shot, avg@322025.12 | 86.75 | — | — | |
| DAPO-FTBackbone=Qwen3-8B, Evaluation Protocol=zero-shot, avg@322025.12 | 84.11 | — | — | |
| DAPOBackbone=Qwen3-8B, Evaluation Protocol=zero-shot, avg@322025.12 | 83.96 | — | — | |
| PPPOBackbone=Qwen3-4B, Evaluation Protocol=zero-shot, avg@322025.12 | 83.06 | — | — | |
| DAPO-FTBackbone=Qwen3-4B, Evaluation Protocol=zero-shot, avg@322025.12 | 82 | — | — | |
| DAPOBackbone=Qwen3-4B, Evaluation Protocol=zero-shot, avg@322025.12 | 81.63 | — | — | |
| GRPOBackbone=Qwen3-8B, Evaluation Protocol=zero-shot, avg@322025.12 | 79.44 | — | — | |
| INTUITORBackbone=Qwen3-8B, Evaluation Protocol=zero-shot, avg@322025.12 | 78.24 | — | — | |
| GRPOBackbone=Qwen3-4B, Evaluation Protocol=zero-shot, avg@322025.12 | 76.77 | — | — | |
| INTUITORBackbone=Qwen3-4B, Evaluation Protocol=zero-shot, avg@322025.12 | 75.83 | — | — | |
| Qwen3-8BBackbone=Qwen3-8B, Evaluation Protocol=zero-shot, avg@322025.12 | 75.08 | — | — | |
| Qwen3-4BBackbone=Qwen3-4B, Evaluation Protocol=zero-shot, avg@322025.12 | 72.67 | — | — | |
| SENTBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 70 | 71.56 | 70.16 | |
| w/ MaskBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 69.61 | 70 | 68.28 | |
| w/ EnBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 65.47 | 65.62 | 63.75 | |
| w/ ClipBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 65.23 | 64.49 | 67.03 | |
| w/ High EnBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 64.69 | 60 | 62.97 | |
| w/ CovBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 64.38 | 60.94 | 65.94 | |
| DAPOBackbone=Qwen3-1.7B, Evaluation Protocol=zero-shot, avg@322025.12 | 63.03 | — | — | |
| PPPOBackbone=Qwen3-1.7B, Evaluation Protocol=zero-shot, avg@322025.12 | 62.89 | — | — | |
| DAPO-FTBackbone=Qwen3-1.7B, Evaluation Protocol=zero-shot, avg@322025.12 | 62.8 | — | — | |
| INTUITORBackbone=Qwen3-1.7B, Evaluation Protocol=zero-shot, avg@322025.12 | 60.99 | — | — | |
| GRPOBackbone=Qwen3-1.7B, Evaluation Protocol=zero-shot, avg@322025.12 | 60.24 | — | — | |
| GRPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 58.12 | 56.25 | 60.47 | |
| w/ AdvBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 58.12 | 60.31 | 56.41 | |
| Base modelBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.12 | 57.58 | 57.81 | 57.03 | |
| Qwen3-1.7BBackbone=Qwen3-1.7B, Evaluation Protocol=zero-shot, avg@322025.12 | 56.93 | — | — | |
| CISPOModel=Qwen3-14B-Base2026.02 | — | — | 89.84 | |
| CISPOModel=Qwen3-30B-A3B-Base (MoE)2026.02 | — | — | 87.03 | |
| CISPOModel=Qwen3-8B-Base2026.02 | — | — | 80.94 | |
| DAPOModel=Qwen3-14B-Base2026.02 | — | — | 87.66 | |
| DAPOModel=Qwen3-30B-A3B-Base (MoE)2026.02 | — | — | 83.75 | |
| DAPOModel=Qwen3-8B-Base2026.02 | — | — | 80.94 | |
| DISPOModel=Qwen3-14B-Base2026.02 | — | — | 92.03 | |
| DISPOModel=Qwen3-30B-A3B-Base (MoE)2026.02 | — | — | 89.38 | |
| DISPOModel=Qwen3-8B-Base2026.02 | — | — | 87.66 | |
| Starting checkpointModel=Qwen3-14B-Base2026.02 | — | — | 22.19 | |
| Starting checkpointModel=Qwen3-30B-A3B-Base (MoE)2026.02 | — | — | 12.5 | |
| Starting checkpointModel=Qwen3-8B-Base2026.02 | — | — | 24.69 |